ValueError: Shapes (None, 27) and (None, 1) are incompatible

X = df.body
y = df.target

num_classes = len(y.unique())

tokenizer = BertTokenizer.from_pretrained('DeepPavlov/rubert-base-cased')
X_enc = tokenizer(X.tolist(), padding=True, truncation=True, max_length=100, return_tensors='np')

X_train, X_val, y_train, y_val = train_test_split(X_enc['input_ids'],
                                                  y,
                                                  random_state = 42,
                                                  stratify = y,
                                                  test_size = 0.3)
X_train = tf.convert_to_tensor(X_train)
X_val = tf.convert_to_tensor(X_val)

y_train = np.array(y_train)
y_val = np.array(y_val)

model = TFBertForSequenceClassification.from_pretrained('DeepPavlov/rubert-base-cased', num_labels=num_classes, from_pt = True)

model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=5e-5),
              loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=[tf.keras.metrics.AUC(multi_label=True)])
model.fit(X_train, y_train, epochs=3, batch_size=32)

Код падает на последней строке с ValueError.

Значения размерностей тысячу раз перепроверены:

X_train shape: (39522, 100) 
y_train shape: (39522,)  
X_val shape: (16938, 100) 
y_val shape: (16938,) 
X_train type: <class 'tensorflow.python.framework.ops.EagerTensor'> 
y_train type: <class 'numpy.ndarray'>